A reliability- and latency-driven task allocation framework for workflow applications in the edge-hub-cloud continuum

A team of researchers from the University of Cyprus has developed a breakthrough optimization framework that dramatically improves how AI workflow applications run across distributed computing environments. Their paper, 'A reliability- and latency-driven task allocation framework for workflow applications in the edge-hub-cloud continuum,' addresses a critical challenge in modern AI deployment: efficiently distributing computational tasks across edge devices, hub devices, and cloud servers while balancing competing objectives of reliability and latency.

The framework employs an exact multi-objective optimization approach using binary integer linear programming (BILP) that considers the relative importance of each objective. Unlike existing methods, it holistically addresses device limitations, diverse operating conditions, and incorporates time redundancy techniques while accounting for crucial constraints often overlooked in related studies. In testing with a real-world workflow application, the method achieved remarkable improvements—84.19% better reliability and 49.81% lower latency compared to baseline strategies across relevant objective trade-offs.

This research matters because critical AI applications—from autonomous vehicles to industrial IoT systems—increasingly rely on streamlined edge-hub-cloud architectures rather than conventional edge computing alone. The framework's scalability is demonstrated across diverse workflow structures, sizes, and criticality levels, with practical runtimes averaging from 0.03 to 50.94 seconds. The work has been accepted for publication in Future Generation Computer Systems (FGCS), indicating its significance for the distributed computing community and practical implications for deploying reliable, low-latency AI systems in real-world scenarios.